问答系统中问题模式分类与相似度计算方法  

Methods of questions pattern classification and similarity measure for question answering system

作　　者：周建政 谌志群[2] 李治 王荣波[2] 冯凯[2] 

机构地区：[1]天格科技(杭州)有限公司,杭州310005 [2]杭州电子科技大学认知与智能计算研究所,杭州310018

出　　处：《计算机工程与应用》2014年第1期116-120,共5页Computer Engineering and Applications

基　　金：杭州市科技发展计划重大科技创新专项(No.20122511A18);国家自然科学基金青年项目(No.61202281)

摘　　要：基于FAQ库的限定域自动问答系统由于更具实用性而成为自然语言处理领域的研究热点,而问题之间的相似度计算是其中最关键的技术。现有的问句相似度计算技术在处理带有上下文情景描述的问题时效果较差。针对现有技术存在的问题,提出将用户问题分为简洁模式问题(SMQs)和情景模式问题(CMQs),并提出了基于规则的问题模式分类算法。在此基础上,进一步提出了综合考察情景相似度和问句相似度的情景模式问题(CMQs)相似度计算方法。实验结果表明,问题模式分类算法取得了90%以上的准确率和召回率,情景模式问题相似度计算方法在时间复杂度较低的情况下也取得了74.3%的正确率。At present, question answering system based on Frequently Asked Questions（FAQ）for restricted domains is a research focus in the field of natural language processing due to its practicality. The similarity measure between questions plays a very important role in one question answering system. The traditional questions similarity measure technologies have unsatisfactory effects for those questions with context information. A rule-based question pattern classification algo-rithm is proposed for dividing all questions into two categories：Simple Mode Questions（SMQs）and Context Mode Ques-tions（CMQs）. Then, a similarity measure method for CMQs is presented in which the similarities between context infor-mation and that between questions are combined together. The experimental results show that both precision and recall rate of the proposed question pattern classification method exceed 90%, and the accuracy of similarity measure for con-text mode questions reaches 74.3%with lower time complexity.

关 键 词：相似度计算 模式分类 上下文信息 问答系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]